Abstract:Large-eddy and direct numerical simulations generate vast data sets that are challenging to interpret, even for simple geometries at low Reynolds numbers. This has increased the importance of automatic methods for extracting significant features to understand physical phenomena. Traditional techniques like the proper orthogonal decomposition (POD) have been widely used for this purpose. However, recent advancements in computational power have allowed for the development of data-driven modal reduction approaches. This paper discusses four applications of deep neural networks for aerodynamic applications, including a convolutional neural network autoencoder, to analyze unsteady flow fields around a circular cylinder at Re = 100 and a supersonic boundary layer with Tollmien–Schlichting waves. The autoencoder results are comparable to those obtained with POD and spectral POD. Additionally, it is demonstrated that the autoencoder can compress steady hypersonic boundary-layer profiles into a low-dimensional vector space that is spanned by the pressure gradient and wall-temperature ratio. This paper also proposes a convolutional neural network model to estimate velocity and temperature profiles across different hypersonic flow conditions.

What problem does this paper attempt to address?

This paper aims to solve the problems of data dimension reduction and pattern decomposition in complex flow phenomena in fluid dynamics. Specifically, researchers attempt to use deep neural networks (especially convolutional neural network auto - encoders) to perform reduced - order modeling of steady - state and unsteady - state flows. This includes: 1. **Data Dimension Reduction and Pattern Decomposition**: - The paper explores how to use deep neural networks (especially convolutional neural network auto - encoders) to perform dimension reduction and pattern decomposition on flow field data. Traditional pattern decomposition methods such as proper orthogonal decomposition (POD) and spectral proper orthogonal decomposition (SPOD) have limitations when dealing with nonlinear flow phenomena, while deep learning methods can better capture these nonlinear features. 2. **Analysis of Application Cases**: - **Cylinder Wake**: Researchers analyzed the unsteady flow in the two - dimensional cylinder wake with a Reynolds number of 100, and compared the coherent structures identified by the auto - encoder, POD, and SPOD methods. - **Supersonic Boundary Layer**: The supersonic boundary layer flow containing Tollmien - Schlichting waves was studied, and the results of the auto - encoder and POD methods were compared. - **Hypersonic Boundary Layer**: The hypersonic steady - state boundary layer flows under different pressure gradients and wall - temperature ratios were studied. The auto - encoder was used to compress these flow profiles into a low - dimensional vector space, and a convolutional neural network model was proposed to estimate the velocity and temperature profiles under different hypersonic flow conditions. 3. **Model Performance Evaluation**: - By comparing the performance of the auto - encoder and traditional methods (such as POD and SPOD), the paper shows the advantages of the auto - encoder in data compression and pattern representation. Especially in the case of the cylinder wake, the error of the flow field reconstructed by the auto - encoder is significantly lower than that of the POD method, indicating that the auto - encoder can better capture nonlinear flow features. 4. **Conclusion**: - The research results show that the auto - encoder method based on deep neural networks performs well in identifying dominant flow patterns (especially large - scale coherent motions) and is suitable for constructing reduced - order models. For large - scale data sets or flows dominated by strong nonlinear phenomena, the auto - encoder method is a very attractive alternative because it does not need to solve large - scale singular value problems and can be designed to capture nonlinear features. - However, one drawback of the auto - encoder method is that it requires a large amount of training data. In general, through multiple case studies, this paper verifies the effectiveness and superiority of deep neural networks in fluid - dynamic data dimension reduction and pattern decomposition.

Reduced-Order Modeling of Steady and Unsteady Flows with Deep Neural Networks

A New Data Assimilation Method of Recovering Turbulent Mean Flow Field at High Reynolds Numbers

Deep neural networks for nonlinear model order reduction of unsteady flows

Deep Learning Methods for Reynolds-Averaged Navier-Stokes Simulations of Airfoil Flows

Data-driven reduced order modeling of environmental hydrodynamics using deep autoencoders and neural ODEs

Data-driven reduced-order modeling for nonlinear aerodynamics using an autoencoder neural network

Assessment of unsteady flow predictions using hybrid deep learning based reduced-order models

Combined space-time reduced-order model with 3D deep convolution for extrapolating fluid dynamics

Deep Convolutional Recurrent Autoencoders for Flow Field Prediction

Combined space–time reduced-order model with three-dimensional deep convolution for extrapolating fluid dynamics

Reduced-order Model for Fluid Flows via Neural Ordinary Differential Equations

Reduced-order Deep Learning for Flow Dynamics

An accuracy-enhanced transonic flow prediction method fusing deep learning and reduced-order model

Deep learning-based predictive modelling of transonic flow over an aerofoil

Application of Deep Learning Method to Reynolds Stress Models of Channel Flow Based on Reduced-Order Modeling of DNS Data

Reduced-order modeling for stochastic large-scale and time-dependent flow problems using deep spatial and temporal convolutional autoencoders

Time-variant prediction of flow over an airfoil using deep neural network

Machine-Learning for Nonintrusive Model Order Reduction of the Parametric Inviscid Transonic Flow past an airfoil

Data-driven modeling of unsteady flow based on deep operator network

Steady-State Transonic Flowfield Prediction via Deep-Learning Framework

Fast Transonic Flow Prediction Enables Efficient Aerodynamic Design